Johan Ugander

The effects of Facebook and Instagram political advertisements on the 2020 US election Nature Human Behaviour, Published online: 05 March 2026; doi:10.1038/s41562-025-02329-9To study the effects of political advertising, we conducted a field experiment with over 60,000 participants. Removing political advertisements from the Facebook and Instagram feeds of randomly selected participants before the 2020 US election did not have a detectable effect on political knowledge, polarization, turnout or political participation.

The effects of Facebook and Instagram political advertisements on the 2020 US election

Worthwhile new essay "Mathematicians in the Age of AI" by Jeremy Avigad, CMU professor and director of the NSF Institute for Computer-Aided Reasoning in Mathematics (ICARM) at CMU: www.andrew.cmu.edu/user/avigad/...

Harsh Parikh will present work on how to transport estimated effects from one set of networks to another.

Shuangning Li will share new results on covariate adjustment in experiments.

So if you're working on networks+causality, consider participating. Abstracts due March 10th causnets.github.io

Looking forward to this!

The Median Voter Theorem is a Clarity Trap What the Democratic party needs; what it demands, is bold, persistent experimentation

The Median Voter Theorem is a Clarity Trap www.programmablemutter.com/p/the-median...

We've got 4 great invited speakers for our satellite event causnets.github.io at @netsciconf.bsky.social.
causnets.github.io/speakers/

In two nicely related talks, Christina Lee Yu & @vivianodavide.bsky.social will each present work on cluster-randomized designs in networks.

Valhallavägen? iykyk

We're all connected. But how...exactly?

A quick dip into the history of network analysis in social science.

I got frustrated copying quotes from PDFs with line breaks, and used Claude to make this little tool: mimno.github.io/copyoneline/

Paste text into the box, it removes newlines and puts the result back in your clipboard, adding quotation marks if desired.

My n=1 experience with LLM survey responses (for piloting) is that they are bad at representating population-level distributions of opinions, even when one plays with temperature settings. Very cool manuscript that drills in to joint distributions. (paper link at end of thread)

Online opt-in surveys also find recent religious resurgence among U.S. young adults While this analysis focuses on claims of religious revival among young adults in the U.K., some opt-in surveys have pointed to a similar trend in the United States. Barna Group, a research organization serving Christian leaders, has used online opt-in survey data to make claims of rising churchgoing among young adults in the U.S. According to Barna, “Since 2019, both Gen Z and Millennials were the least likely generation to frequently attend church. Today, they are the most engaged.” However, surveys from Pew Research Center using random samples show no clear evidence of a religious revival among young adults. Nor is there clear evidence of religious revival in two other surveys based on random samples conducted by other organizations: the General Social Survey and the American Time Use Survey.

Is there a revival of churchgoing among US young adults? According to
Opt-in online polls: Yes
Surveys using random samples of the population: No
https://www.pewresearch.org/short-reads/2026/01/23/has-there-been-a-christian-revival-among-young-adults-in-the-uk-recent-surveys-may-be-misleading/

🎯

Tired: game theory of nuclear non-proliferation
Wired: game theory of chatbot war games
Inspired: game theory of why did a journalist cover this paper, why is this media story in my feed

I'm hiring a postdoc at @cmu.edu (w/ far.ai & @dgrand.bsky.social + @gordpennycook.bsky.social)!

How do LLMs shape human beliefs — and what do we do about it? AI safety meets behavioral science.

Open to technical and social science backgrounds.

Twitter: Last Week Tonight with John Oliver (HBO) YouTube video by LastWeekTonight

John Oliver and team do a great job here of explaining how Twitter/X is a cesspool and why you shouldn’t go anywhere near it. Only thing it misses is new research showing that its right-biased algorithm “works” in that it shifts users’ views rightward. www.youtube.com/watch?v=p7ZG...

You know that annoying NSF form "List every coauthor/co-PI from the last 4y" ?

At @cevianlabs.io we built a free tool that drafts the COI form from your PDF CV in minutes. Check it out 👇

Living the metascience dream (or nightmare) with AI for science What happens when we go from replication crisis to robustness extremes?

AI makes continuous reproducibility and robustness testing trivial. What happens to science under new levels of scrutiny and stress-testing by default?

Some thoughts on how this could play out, informed by watching open science play out over the last decade.

Thanks for sharing. This very much matches my experiences. These tools are particularly good at porting and making code faster, especially if there are unit tests in place to track consistent validity. It's wild.

"…while working on this project I thought of recent comments I’ve heard such as 'Subject X will be the first to go'. This is folly. Nothing is going anywhere. Science is about to get better and it will proceed faster than at any time in history."

Can feed algorithms shape what people think about politics? Our paper "The Political Effects of X's Feed Algorithm" is out today in Nature and answers "Yes."

www.nature.com/articles/s41...

CausNetS: Toward a Causal Network Science A NetSci 2026 Satellite

⚙️ Working at the intersection of causality and networks?

We're organizing a satellite event at @netsciconf.bsky.social in Boston on June 1st. The focus is networks science and causal inference.

Submit your work by March 10th!

causnets.github.io

Listening to My Students at Scale: Exit Tickets, NotebookLM, and the Tightest Feedback Loop I've Ever Built It started at a teaching workshop, last semester: Craig Kapp and Rob Egan presented a seminar at the NYU Center for Teaching and Learning ...

Panos Ipeirotis (NYU): "How I started using exit tickets (short feedback surveys after every class) and processing them through NotebookLM to generate follow-up materials before the next session. The feedback loop went from one semester to a few hours." www.behind-the-enemy-lines.com/2026/02/list...

I've had very similar experiences with Claude/Codex the last few weeks. Vibe-code a simple python implementation, vibe-spec some unit tests, and then vibe-optimize. Parallelizing simulation code? Port python to Cython? So far both Claude and Codex frontier models have been flawless at it. 3/3

I specifically remember an assignment on matrix multiplication, and how hard it was to get Strassen's algorithm to run faster than a well-optimized O(n^3) algorithm—at the scale of 512x512 matrices—largely because it was much easier to optimize how the O(n^3) algorithm did caching. 2/

When I was taught about compiler optimization in undergrad ("what does gcc -o3 do?") we were taught a bunch of perf tricks (loop unrolling, etc.), but a key lessons was: rather than write tricky code, it's often better to write simple, correct code and then let the optimizer do it's thing. 1/

A line curve showing number of awards for fiscal year 2026 compared to fiscal years 2021-2025 across NSF. The fiscal year 2026 curve lies well below curves for other fiscal years.

NSF Update

Funding curve overall. A little bit of progress in the past week, but only a little bit.

Now by Directorate...

1/11

Cryofront The Pareto frontier of sustained cold — explore minimax temperature curves by city or ZIP code.

I've update the cryofront tool I built yesterday (for visualizing sustained cold spells) to now also visualize "thermofronts," sustained warm spells. Also some new features like supporting non-US cities, toggling today's data on/off, etc. jugander.github.io/cryofront/

Teletherm When does your hottest and coldest day fall? Track how the timing of annual temperature extremes drifts over 75 years.

Ah! The Teletherm link got garbled: jugander.github.io/cryofront/te...

And yes, the hot nights version is cooking, will probably get it up tomorrow :)

Gah! Bsky autocomplete failed me. Fixed and reposted!

I asked Claude to cook a port of this code to visualize "teletherms", the idea from the 2016 paper by
@peterdodds.bsky.social @lewismath.bsky.social @andyreagan.com @chrisdanforth.bsky.social. The zero-shot implementation (of my take on it) was nearly flawless. jugander.github.io/cryofront/te...